Fatigue characterization of an asphalt mixture is commonly estimated by laboratory fatigue tests. Based on the classical fatigue analysis, fatigue lives obtained from different test devices are not comparable even when they are performed at the same test conditions. It is believed that there are two main reasons causing the difference in fatigue results being the difference in stress-strain distribution of the different specimens and the fatigue analysis approach. This research focuses on the harmonization of the fatigue results obtained from the different methods, which are recommended by the European standard EN 12697-24. The main goal is to find a correlation between the different fatigue test methods and to improve the classical fatigue analysis approach to better represent the actual fatigue characteristics of asphalt mixtures. To realize the main objective of the study, an extensive fatigue testing program was carried out on dense asphalt concrete 0/8 (DAC 0/8). In the program, uniaxial tension and compression (UT/C) fatigue tests, four-point bending (4PB) fatigue tests and indirect tensile (IT) fatigue tests were performed. For each fatigue test, specimens with different sizes were tested to explore the size effect on the fatigue results. In order to limit the test program, the tests were performed in two loading modes, at two temperatures and one frequency. For the 4PB fatigue test, the measured displacement highly depends on the properties of the loading frame. Calibration tests on the 4PB test setup were conducted to obtain the pure bending deflection of the beam. Comparison of the fatigue results obtained with the different test methods at the same test condition and loading mode shows that the fatigue life from the 4PB test is the longest and from the IT test is the shortest. Because of the homogeneous tensile strain field, the UT/C and IT fatigue results are not significantly influenced by the specimen size. However, the 4PB test results depend on the dimension of the used specimen, because the stress-strain field of the beam specimen varies along the length and cross section. The partial healing (PH) model was used to determine the relationship between the UT/C and 4PB fatigue results in strain-controlled mode. It is a material model that describes the evolutions of the stiffness and phase angle for a unit volume during testing. This implies that the model can be directly applied to the fatigue results obtained from the “homogenous” tests, the UT/C test, because the strain is uniformly distributed throughout the specimen. When analyzing the 4PB results, the backcalculated stiffness is not the local stiffness of the material but the so-called weighted overall stiffness of the whole specimen. Therefore a weighing procedure is required to calculate the weighted overall stiffness modulus from the local stiffness by taking into account the dimensions and the strain distribution of the specimen. By adjusting the model parameters, the PH model provides a good simulation for the evolutions of the stiffness and phase angle. All the model parameters can be expressed as functions of the applied strain level. The trends of the parameter ??1 and ??2 indicate the existence of an endurance limit. The predicted endurance limit obtained by the UT/C test is around 68 ?m/m for the tested DAC 0/8 mixture, which does not change with specimen size and temperature. For the 4PB fatigue test, the local stiffness at different parts of the beam can be calculated by means of the PH model. The evolution of the calculated stiffness at the surface in the midsection of the beam is comparable with the UT/C fatigue results when the pure bending strain on the beam surface is equal to the tensile strain in the cylinder. Therefore the PH model offers a possibility to compare different fatigue results. In the second part of this research, the yield surface concept was applied to develop a new fatigue analysis approach. As a visco-elastic material, the yield surface of an asphalt mixture highly depends on temperature and strain rate. Therefore, monotonic uniaxial compression (MUC) and monotonic uniaxial tension (MUT) tests were performed at different temperatures and strain rates to derive such yield surfaces. For the three fatigue tests, the yield surface at the critical location of the specimen was determined in the I1-?J2 space. The fatigue results were then interpreted by comparing the actual stress condition with the yield surface. A new parameter R? was introduced as an indicator of the “safety against failure”. By comparing the R? values at the different locations of the specimen for the IT test, the weakest points are found at the locations with the maximum horizontal tensile strain, which are close to the loading strips, instead of the center of the specimen. A straight line was found by plotting R? at the critical location and the fatigue life on a log-log scale. Compared to the traditional fatigue analysis, the size effect on the fatigue results was excluded by using this new fatigue relation. For the stress-controlled mode, the fatigue lines obtained from the UT/C test show a good agreement with the IT fatigue results. Of course the influence of temperature and loading mode still exists in this new fatigue method. In the normal coordinate, when the fatigue life tends to infinity, R? becomes a constant value, denoted by Rlimit. The parameter Rlimit represents the endurance limit in a three-dimensional state. This value does not change with specimen size and test type but is influenced by the temperature and loading mode. Based on all the test results and their analysis in this research, it was concluded that the UT/C fatigue results are material properties and not influenced by specimen size. The PH model provides a good simulation of fatigue behavior of the asphalt mixture and is able to find the correlation between the UT/C and 4PB test results. The developed fatigue analysis approach characterizes the fatigue performance of asphalt mixtures in three-dimensional state, which is more close to the field situation. The endurance limit predicted by the new fatigue approach is independent of the specimen size and fatigue test type.